Active shape models (ASMs) are statistical, deformable models, exhibiting a remarkable performance for the segmentation of the lung fields in plain chest radiographs. In this paper we propose a novel approach to improving the robustness of the original ASM against weak lung field boundaries, which can cause leaking of the shape's contour into the lung fields. The ASM is shielded against leaking by the prior application of a grey-level selective thresholding scheme that subtracts irrelevant anatomic structures from the radiograph. The proposed approach copes with affine lung field projections and features resistance to the presence of dense external objects used for patient's monitoring and support. Its advantageous performance is demonstrated on a challenging set of chest radiographs obtained from patients with bacterial pulmonary infections.